LangChain Tests a New Judge, Notion Builds Agent Memory, and Peter Sellis on What Makes a Great PM

A quieter Saturday, but three things worth your attention: LangChain tested a new System One model as an agent evaluator and published the results (including where it fails), Notion shipped a technical look at how they're building persistent shared memory for AI agents, and Lenny's conversation with Peter Sellis surfaces some of the sharpest PM craft thinking I've seen in a while.